# Counting irrigated fields and center pivots in the United States from open data

**A per-field accounting over 22.5 million satellite-mapped field boundaries**

UFFDA · Working draft, 2026-07-30 · Shared for expert review

---

## Summary of findings

- **Irrigated extent.** Persistent irrigation covers roughly 52–55 million acres of the conterminous United States (52.2M acres on the ≥8-of-21-mapped-years basis, 1997–2017; 55.5M acres on the full ≥8-of-24-mapped-years basis, 1997–2020; 2024 boundary vintage) — within a few percent of all three federal anchors: USDA's 2023 Irrigation and Water Management Survey (53.1M acres), the 2022 Census of Agriculture (54.9M acres), and USGS MIrAD-US (about 54.4M acres).
- **Irrigated field count.** At a ≥10-acre floor, 565,699 field polygons meet the ≥8-of-21-years persistence bar (942,358 with no floor; 1,012,716 with no floor on the 24-year basis). No official count of irrigated fields exists; two independent triangulations — 2.66 fields per irrigated farm against USDA's 212,714 irrigated farms, and roughly 581,000 implied by scaling the Yan & Roy (2016) national crop-field census by the irrigated share of cropland — corroborate the ≥10-acre figure.
- **Center pivots.** An estimated 160,000–210,000 pivot-equivalent circular irrigated fields nationally — a range anchored on the USDA-2023-implied bracket (25.97M center-pivot acres ÷ an assumed 125–160 acres per system = 162,000–208,000) — of which at least 58,890 are individually confirmed by overlap with an independently mapped pivot polygon (a floor, not a ceiling). These are counts of land parcels, never of machines.
- **State patterns.** Texas leads the nation in pivot-overlap field-polygon count (70,223) while Nebraska leads in center-pivot acres (6.70M, USDA 2023) — the count-up, acres-down pattern that declining well capacity on the southern High Plains aquifer predicts; a boundary-segmentation confound in the Texas panhandle is flagged and untested.
- **Size spectrum.** 59% of all mapped field polygons are under 1 acre yet carry 0.49% of mapped acreage — a segmentation floor artifact, not real fields — while persistent-irrigation acreage peaks exactly in the 80–160-acre quarter-section bin (28.2% of irrigated acres). This is why every count published here states its area floor.

---

## 1. The question, and why no official answer exists

How many irrigated fields are there in the United States? How many of them are center pivots?

Neither question has an official answer. USDA measures irrigation in **farms** and **acres**: the 2023 Irrigation and Water Management Survey (IWMS) reports 212,714 farms irrigating 53.1 million acres, of which 25.97 million acres are under center-pivot systems. The 2022 Census of Agriculture reports 54.9 million irrigated acres. No federal instrument counts **fields**, and none counts **irrigation machines**. Remote-sensing irrigation products — LANID (the Landsat-based Irrigation Dataset), MIrAD-US (the MODIS Irrigated Agriculture Dataset for the US), and LGRIP30 (the Landsat-derived Global Rainfed and Irrigated-Cropland Product) — map extent, pixels and acres, not field-level units. Industry figures for installed pivot machines exist but are dated, continental in scope, or both.

This paper describes a per-field accounting built from open data: an attribute table over every satellite-mapped field-boundary polygon in the conterminous US, joined against per-pixel irrigation histories and an independently mapped center-pivot polygon layer. The result is a set of bracketed estimates with stated definitions, each checked against the published federal and academic record.

## 2. Approach

We ran one batch enrichment pass over the **Fields of the World (FTW) Global** predicted field-boundary dataset for the conterminous US: 22,544,745 polygons in the 2024 model vintage and 24,857,861 in 2025 (47.4M polygon-vintage rows across 1,285 one-degree tiles; roughly 5.3 hours of compute on a single workstation). Each polygon received a set of per-field columns: geodesic geometry metrics, irrigation-history statistics (LANID), a single-epoch irrigation class fraction (LGRIP30), county assignment, and center-pivot overlap (GCPIS — the independently mapped pivot-polygon layer described in Section 3). The 2024 vintage is the reporting basis throughout; all headline quantities are stable within 1–4% against the 2025 vintage.

**Geodesic-only measurement.** All areas and perimeters are computed on the WGS84 ellipsoid from the native longitude/latitude geometry. No Web Mercator projection appears anywhere in the measurement path: Mercator area inflation grows as 1/cos²(latitude) — roughly 1.79× at Corn Belt latitudes — and is a silent-error class in agricultural acreage work.

**Coverage-fraction zonal statistics.** Raster signals are extracted as partial-pixel coverage fractions rather than per-pixel majority or all-touched counts. Field polygons — especially near-quarter-section pivot parcels — have high edge-to-area ratios at 30 m pixel scale, and the quantity of interest is the fraction of a field carrying a signal, not a binary vote. Per-field LANID statistics are coverage-weighted for the same reason.

**One pass, additive columns.** The run was interrupted twice by infrastructure faults and resumed without data loss, finishing with 0 unresolved (tile, source) units.

**What we publish, and what we don't.** Every derived signal is carried as a measurement or hedged score, never a bare verdict; every published quantity states its source, vintage, definition, and area floor; and each headline number was checked against the published record (Section 5) by a reviewer working separately from the analysis, before appearing here. Numbers that failed that review — including several that are computable from our own table — are deliberately absent.

## 3. Data sources

| Dataset | Role | Resolution | Vintage | License / attribution |
|---|---|---|---|---|
| Fields of the World (FTW) Global — predicted field boundaries | The polygon universe being counted | Polygons predicted from 10 m Sentinel-2 imagery | 2024 and 2025 model vintages | CC-BY 4.0 — Taylor Geospatial and collaborators (Kerner Lab / Arizona State University, Microsoft AI for Good Research Lab, Washington University in St. Louis, Oregon State University, Clark University); source.coop/ftw/global-data |
| LANID — Landsat-based Irrigation Dataset, v3 (Xie & Lark) | Annual irrigation maps; the persistence basis | 30 m | 1997–2017 (21 annual maps) | CC-BY 4.0 (Zenodo, DOI 10.5281/zenodo.5548555) |
| LANID 2018–2020 extension (USGS; Martin et al. 2024) | Extends the per-field record to 2020 | 30 m | 2018–2020 (3 annual maps) | Public domain (USGS data release, DOI 10.5066/P9YWR0OJ) |
| LGRIP30 — Landsat-derived Global Rainfed and Irrigated-Cropland Product, V002 (NASA; Teluguntla et al.) | Corroborating single-epoch irrigation signal | 30 m | Nominal 2020 (2019–2021 Landsat record) | Openly shared without restriction per NASA EOSDIS Data Use and Citation Guidance |
| GCPIS — global center-pivot polygons, from GMIE, the Global Maximum Irrigated Extent dataset (Tian et al. 2025) | Independent pivot-footprint layer for the overlap join | Polygons from March–August 2020 imagery | 2020 | CC0 1.0 (Harvard Dataverse, DOI 10.7910/DVN/HKBAQQ) |
| US Census TIGER/Line boundaries | State/county assignment for aggregation | Vector | Administrative | Public domain; acknowledgement to the U.S. Census Bureau |
| USDA NASS: 2022 Census of Agriculture; 2023 IWMS; 2018 Farm and Ranch Irrigation Survey (FRIS) | Benchmark anchors (farms, acres, center-pivot acres) | Tabular | 2017–2023 | US federal public domain |
| USGS MIrAD-US — Moderate Resolution Imaging Spectroradiometer (MODIS) Irrigated Agriculture Dataset for the US | Benchmark anchor (irrigated extent) | 250 m | Multi-epoch, latest 2017 | US federal public domain |

All inputs are open; the two CC-BY sources are attributed in full in the footer. LANID and LGRIP30 cover the conterminous US, so every irrigation result here is conterminous-US in scope.

## 4. Definitions

**Persistently irrigated: ≥8 of the mapped years.** LANID provides 24 annual irrigation maps spanning 1997–2020 (21 from the v3 release, 3 from the USGS extension). A field polygon counts as persistently irrigated when its coverage-weighted years-irrigated total reaches at least 8 of the mapped years. The threshold comes from a validation sweep, not convention: county-level agreement with USDA data is non-monotonic in the threshold (requiring more years does not keep improving agreement), and ≥8 performed best. Where a result predates the 2018–2020 ingest, it is labeled with its 21-year basis (≥8 of 21, 1997–2017); the full-record basis is ≥8 of 24 (1997–2020). Extending the window mechanically adds fields that cross the threshold with the three new years (+7.5% fields, +6.3% acres); that shift is a definition-window effect, not evidence of irrigation growth — USDA's own record shows national irrigated area *declining* about 5% over 2017–2023.

**Why persistence, not a snapshot.** The cautionary result is measured, not hypothetical: LGRIP30's single-epoch irrigated class, applied alone, overcounts USDA county irrigated acreage by a median of +102%, and by up to +1,123% in humid counties, where any false-positive signal lands on a small true denominator. A single-epoch classifier cannot distinguish irrigated land from land that merely looked irrigated in one wet-need window. We use LGRIP30 only as a corroborating recent-epoch signal — carried as "irrigation signal in the 2019–2021 Landsat record," never as "irrigated" — and build every headline on multi-year persistence.

**Area floors, stated with every count.** The boundary product contains a sub-acre artifact population (Section 5.5): 13.32M polygons under 1 acre — 59.1% of all polygons — carrying 0.49% of mapped acreage, with the tightest artifact band (620,813 polygons at roughly 0.02 acres), where 100% of a vertex-count sample had the vertex signature of a simple rectangle. Floors are applied at query time, never baked into the table, and every published count names its floor. The default reporting floor is ≥10 acres, matching US field-size domain expectations. Acreage totals are nearly floor-insensitive (52.19M acres with no floor → 49.98M at ≥20 acres, −4.2%), which is why acreage claims are more robust than count claims throughout.

**Polygon counts, not machines.** Every count here is a count of satellite-mapped field-boundary polygons (or, where stated, of pivot-layer polygons). One physical pivot can straddle two boundary polygons; one pivot-layer polygon can represent many machines — we observed a single 12,215-acre GCPIS polygon that plausibly aggregates 90+ quarter-section systems. Among persistence-irrigated fields that overlap the pivot layer, 138,075 field polygons form 154,188 intersecting (field, pivot-polygon) pairs — a 1.117:1 ratio, so neither layer's unit maps 1:1 onto physical systems. We never publish a machine count.

**The pivot-overlap bracket.** The boundary model draws parcels, not circles: polygon circularity is capped near π/4 for a square parcel and cannot detect pivots (zero exceptions in a 374k-polygon pilot). The pivot signal is instead an overlap join against GCPIS, an independently mapped global center-pivot polygon layer: per field, the fraction of polygon area covered by pivot polygons. A full circle inscribed in its quarter-section parcel predicts about 0.785 coverage; the observed national distribution among overlapping fields is bimodal (median 0.651; 44.4% below 0.50 — edge brushes; 33.5% at ≥0.90 — full-parcel pivot fields), matching that geometry. The headline range is constructed from the survey side: USDA's 25.97M center-pivot acres (2023 IWMS) divided by an assumed 125–160 acres per system gives 162,000–208,000, published as 160,000–210,000. The size assumption is corroborated in our table: fields at ≥50% pivot overlap average 157 acres.

## 5. Results, with benchmarks side-by-side

| # | Quantity (2024 vintage) | Our value | Benchmark(s) | Benchmark check |
|---|---|---|---|---|
| 1 | Persistent-irrigation acres | 52.19M (≥8 of 21, no floor) / 51.43M (≥10 ac); 55.45M (≥8 of 24, no floor) | USDA IWMS 2023: 53.1M; Census 2022: 54.9M; MIrAD-US: about 54.4M; LANID annual active range: 55.9–61.0M | Within range |
| 2 | Persistently irrigated field polygons | 565,699 (≥10 ac, ≥8 of 21); 942,358 (no floor); 1,012,716 (no floor, ≥8 of 24) | No published counterpart; triangulations: about 2.7 fields/farm implied vs 2.66 computed; about 581,000 via Yan & Roy scaling | Within range (triangulated) |
| 3 | Pivot-equivalent circular irrigated fields | 160,000–210,000 (range); ≥58,890 confirmed floor | NASS-implied 162k–208k; Evans/ARS 2001 about 125k; Valmont 2016 about 265k (North America, incl. lateral-move) | Within range (as a bracket) |
| 4 | Pivot-footprint acres (≥50% overlap) | 14.80M ac (94,074 fields, ≥10 ac) | GMIE's own implied US total about 13.8M ac; Pivot-Net 15.1M ac (2018) | Within about 7% of the source's own total |
| 5 | State pattern | TX leads pivot-overlap count; NE leads pivot acres | NASS 2023: NE 6.70M center-pivot acres, #1 | Outside — explained (hydrogeology), one confound flagged |

**5.1 Irrigated extent.** 52.19M acres (≥8-of-21 basis, no floor; 51.43M at ≥10 acres) sits below every anchor — 1.7% below the 2023 survey figure and up to 14.4% below the top of the annual-active range — the expected signature of a persistence definition being a stricter filter than a single-year survey or annual-active count, not an error. The full 24-year basis gives 55.45M acres, within about 1–4% of the Census and survey anchors. This is the single most externally corroborated number in the analysis: three independent federal/academic instruments land in one tight band around it.

**5.2 Irrigated field count.** There is no published irrigated-field count to compare against — so the check is triangulation. (a) USDA's 212,714 irrigated farms and our 565,699 polygons (≥10 ac) imply 2.66 fields per irrigated farm; independently, USDA's 249.6 irrigated acres per farm divided by our implied 90.9 acres per irrigated field gives 2.74 — agreement within about 3.3%. (b) Yan & Roy's national crop-field census (about 4.18M conterminous-US crop fields, circa 2010), scaled by irrigation's share of total cropland (53.1M of 382M acres, 13.9%), implies roughly 581,000 irrigated fields — the same ballpark as our 565,699. The implied mean irrigated field size (90.9 acres) falls between the all-crop average Yan & Roy report (roughly 48–69 acres) and the 125–160-acre pivot norm, consistent with a blend of large Western pivot fields and smaller Eastern and Californian irrigated parcels.

**5.3 Center pivots.** At least **58,890** are individually confirmed — distinct GCPIS pivot polygons intersecting a persistence-irrigated field — and that figure is a floor, not a ceiling, because some pivot-layer polygons are confirmed aggregates of multiple physical systems. The industry/legacy anchors bracket the range from outside: about 125k machines (Evans/ARS, 2001 — a quarter-century stale floor) and about 265k installed machines (Valmont, 2016 — North-America-wide and including lateral-move and towable units, a geography- and type-mismatched ceiling). The acreage cross-check is stronger than the count check: our ≥50%-overlap pivot-footprint acreage (14.80M acres) reproduces the GMIE source paper's own implied US center-pivot footprint (about 13.8M acres) within about 7%, with a second independently trained model (Pivot-Net, 15.1M acres, 2018) in the same band. Footprint products are a published, acknowledged undercount against USDA survey acreage (Pivot-Net's own comparison runs at 56–58% of the USDA center-pivot figure), which is why the bracket is anchored on the survey side rather than the footprint side.

**5.4 State patterns.** By pivot-overlap polygon count: Texas (70,223) > Nebraska (46,357) > Kansas (29,510) > Colorado (25,022) > Idaho (12,181). By USDA center-pivot acres, Nebraska is first (6.70M, 2023). The inversion has a cited physical mechanism: Nebraska and Kansas sit on thicker, better-recharging sections of the High Plains aquifer, with well yields (>750 gpm) that support full quarter-section circles, while sustained drawdown in the Texas panhandle — declines in saturated thickness exceeding 46 m — forces more numerous, smaller, and partial-arc systems for the same or less total acreage. One confound remains untested: the Texas panhandle is also a documented hotspot for boundary-segmentation artifacts (Section 5.5), so some fraction of the Texas count lead could be over-split polygons rather than physical system density; we flag the claim accordingly rather than presenting it as purely physical. A second pattern is orthogonal to pivots: California leads the nation in persistently irrigated field-polygon count (156,311, ≥8-of-21 basis) yet is absent from the pivot-overlap top ten — California irrigation is predominantly drip, furrow, and flood.

**5.5 The size spectrum.** At the small end, 13.32M polygons (59.1% of all 22.55M) are under 1 acre but hold 0.49% of acreage — a segmentation pixel-floor artifact, not a real smallholding population. The literature supports this reading: at 10 m resolution, only about 28% of genuinely small (sub-0.5 ha) fields survive as distinct polygons ("Fields of the Planet," 2026), so a 10 m-derived product cannot plausibly over-preserve 13 million real sub-acre fields; in a vertex-count sample of the roughly-0.02-acre artifact band, 100% of polygons had the vertex signature of a simple rectangle. At the large end, two distinct populations coexist: 89.6% of >300-acre polygons contain zero pivot polygons (predominantly rangeland/desert mis-segmentation, with single polygons reaching as much as roughly 202,000 acres in Nevada and Oregon — a boundary-QA class, not crop fields), while a real multi-pivot "glob" class of 11,421 polygons (>300 acres, 2+ distinct pivots) — 57.3% of those glob polygons themselves in Nebraska, Kansas, and Texas, the pivot belt — conceals an estimated 21,117 additional pivot-equivalents inside single records. Between the extremes, the distribution behaves as domain priors predict: persistent-irrigation acreage peaks in the 80–160-acre quarter-section bin (28.2% of all irrigated acres, the largest bin by a wide margin), with 160–640 acres contributing another 36.0%.

## 6. Limitations

**County-level distribution is not publishable.** Aggregated to counties, the ≥8-of-21 persistence estimate misses USDA 2022 Census county irrigated acreage: across 2,809 qualifying counties, median absolute error is 96.7%, only 9.4% of counties fall within ±15%, and 78.8% undercount (signed median −86.3%). The mechanism is understood and literature-consistent: humid-region supplemental irrigation is intermittent and rarely clears a strict multi-year persistence bar even when a single-year Census catches it, and part of the gap is a metric mismatch (persistence acres versus a one-year survey). It also means the strong national total benefits partly from offsetting regional errors. Until a regional-first recalibration of the persistence threshold is built and validated, no county-level irrigation claim is published from this table — the national total and the pivot bracket are the defensible results.

**Per-field history ends at 2020.** Nothing newer exists at this resolution: LANID 2018–2020 is the last conterminous-US annual ≤30 m irrigation product; MIrAD-US's latest edition is 2017; IrrMapper covers 11 western states and is not bulk-distributable; ESA WorldCereal offers one global season (2021) at 10 m. The pivot layer is likewise a 2020-imagery epoch, so the whole per-field read coheres on a 2020-era reference joined to 2024/2025 boundaries. The 2021–2025 story is carried only by USDA survey anchors (national irrigated area −5.0 to −5.3% across 2017/2018→2022/2023; center-pivot acres −3.1%), which indicate redistribution — Delta and Southeast growth against western retirement — rather than net national change.

**Boundary-product quality bounds every count.** The FTW boundary layer contains artifacts at both size extremes (Section 5.5): sub-acre segmentation slivers and large-end rangeland mis-segmentation, plus multi-pivot globs in the regions where pivot counting matters most. A sub-field decomposition program — splitting globbed parcels into management units using multi-year crop-rotation evidence and the pivot geometry — was scored case by case in a review kept separate from the build. It failed: 0 of 43 scored cases passed, and the review named seven recurring defect types in the splitting logic. **All decomposition-derived counts are excluded from this paper.** Polygon counts carry boundary-product uncertainty that is proven load-bearing in pivot country, and are framed as boundary-product statistics; acreage aggregates are decomposition-invariant and stand on their benchmark validation.

**Pivot-layer unit looseness.** GCPIS polygons are not machines: confirmed multi-pivot aggregates (the 12,215-acre example), occasional non-circular ghost traces, and tile-grid slicing all exist in the layer. The 58,890 floor is scoped to persistence-irrigated fields, which suppresses ghost-trace inflation, and the aggregate-polygon mechanism biases the floor downward — it remains a floor — but no GCPIS-derived figure should be read as a machine census.

**Texas pivot recall.** In a reviewed Texas panhandle validation stratum, roughly one-third of visibly operating pivots had no GCPIS trace. Texas pivot-overlap counts therefore carry a downward recall bias at the same time as the upward segmentation-artifact confound noted in Section 5.4; the two do not cancel by construction, and the Texas count lead is hedged rather than asserted.

## 7. Reproducibility

Every input dataset is open and cited above with a resolvable DOI or agency source; the benchmarks are public USDA and USGS publications. The method is a single-pass per-polygon enrichment as described in Section 2, with every reported aggregate a query over that table under the definitions stated in Section 4. Methodology detail sufficient for independent reimplementation, and per-number provenance, are available on request through the UFFDA project (uffda.ag).

---

## Attributions

- **Fields of the World (FTW) Global — predicted field boundaries.** Produced by Taylor Geospatial and collaborators (Kerner Lab / Arizona State University, Microsoft AI for Good Research Lab, Washington University in St. Louis, Oregon State University, Clark University). Licensed CC-BY 4.0. https://source.coop/ftw/global-data. Citation: Robinson, C., Muhawenayo, G., Khanal, S., et al. (2026), "The first global agricultural field boundary map at 10 m resolution," arXiv:2605.11055.
- **LANID v3:** Xie, Y. & Lark, T.J. (2021). Landsat-based Irrigation Dataset (LANID), v3 [Data set]. Zenodo. https://doi.org/10.5281/zenodo.5548555. Licensed CC-BY 4.0. (Companion papers: Xie & Lark, *Remote Sensing of Environment*, DOI 10.1016/j.rse.2021.112445; Xie & Lark, *Earth System Science Data*, DOI 10.5194/essd-13-5689-2021.)
- **LANID 2018–2020 extension:** Martin, D., Regan, R.S., Haynes, J.V., Read, A.L., Henson, W., Stewart, J.S., Brandt, J., and Niswonger, R., 2024, LANID: Landsat-based irrigation dataset for CONUS 2018-20: U.S. Geological Survey data release, https://doi.org/10.5066/P9YWR0OJ. US public domain.
- **GCPIS / GMIE:** data files CC0 1.0, Harvard Dataverse, https://doi.org/10.7910/DVN/HKBAQQ. Cited as academic courtesy (not a license requirement): Tian et al. (2025), *Earth System Science Data* 17:855.
- **LGRIP30 V002:** openly shared, without restriction, in accordance with NASA's EOSDIS Data Use and Citation Guidance; citation to Teluguntla et al. (2024) and NASA/USGS/LP DAAC is strongly encouraged and given here.
- **TIGER/Line boundaries:** U.S. Census Bureau, TIGER/Line Shapefiles. Public domain; acknowledgement given per Census Bureau guidance.
- **USDA NASS** (2022 Census of Agriculture; 2023 Irrigation and Water Management Survey; 2018 Farm and Ranch Irrigation Survey) and **USGS MIrAD-US**: works of the US federal government, public domain.

*Benchmark literature cited: Yan & Roy (2016), Remote Sensing of Environment 172:67–86; Tang et al. (Pivot-Net), Remote Sensing 12(3):558; Tian et al. (2025), ESSD 17:855; Xie & Lark (2021), ESSD 13:5689; "Fields of the Planet" (2026), arXiv 2607.04449; USGS Groundwater Atlas HA-730-D; Evans/USDA-ARS (2001); Valmont (2016).*